All Times EDT
Keywords: power prior, adaptive trials, simulation, clinical trial planning
Bayesian statistics have been used for medical device clinical trials in submissions to FDA. Design of Bayesian trials requires careful planning at the design stage. In 2010 FDA finalized the guidance document on the use of Bayesian statistics in medical device clinical trials. It describes two approaches: Bayesian hierarchical modeling and Bayesian adaptive designs. For all studies that employ priors, there needs to be agreement at the outset between the trial designers and the regulatory authorities about its appropriateness. For hierarchical models, the amount of borrowing of the prior depends on how similar the current study is to the prior information from previous trials. If it is very similar, there is a great deal of borrowing and if not there is less or none. In some instances there is a great deal of prior data and then the challenge is to down-weight the prior so it does not overwhelm the current study. This creates a challenge for the design and opens the door for an adaptive approach wherein the amount of borrowing can affect the size if the sample for the current trial. Both power priors and commensurate priors are discussed as well as the combination with propensity scores. In Bayesian adaptive trials usually without informative priors the size of the trial depends on the accumulating data in the trial and so the sample size is not usually fixed in advance, which necessitates the need for simulation under various scenarios at the design stage.